Revolutionary AI Updates Have Landed!

Hello everybody, and welcome back to the Matvid Pro AI channel. Today I have an exciting AI news roundup for you all. And as you can see, yes, my transformation is complete and I have become a real lemon. Just joking, guys. I'm not going to do an entire video like that. However, the AI news portion is truth. There is so much AI news, it's going to be difficult to cover today, but we're going to try.

So.

So let's not waste any more time and get right into the AI news. First up in our AI News roundup, this one is actually a little bit old, but Microsoft's Copilot Studio is getting autonomous agents which are supposed to help with business productivity. Pretty interesting, pretty interesting. As you can see, they're showing off some of the nice UI here. You press new agent or you can import other previously made agents. They have a ton of agents already in here. We've got customer service marketing, product help, case resolution surveys. They seem to be a little bit more basic than maybe some people were hoping, but baby steps, baby steps. You can create these autonomous agents with Copilot Studio and it's going to be in public preview next month. So that's November. So Copilot is your AI assistant. It works for you and Copilot Studio enables you to easily create, manage and connect agents to Copilot. And they're calling agents as the new apps for an AI powered world. I guess that sort of makes sense. Every organization will have a constellation of agents ranging from simple prompt and response to fully autonomous. And again, I don't think the technology is necessarily there yet for like a full autonomous worker replacement. But they're getting better every day. They're going to work on behalf of an individual team or function to execute and orchestrate business processes. And again, we've seen similar things in the past. Think of my Lindy video that I did recently. Also think about the newly announced autonomous agents, the computer use agents and that come from anthropic AI and those look pretty advanced. Again, they're really trying to highlight here that Copilot is how you're going to interact with your agents. It's all going through Microsoft's Copilot, but I guarantee there's going to be autonomous business agent competitors. I think you're going to see a lot of those cropping up in 2025 especially. Anyways, we have a couple of examples here. We have a sales qualification agent enable sellers to focus their time on the highest priority sales opportunities while the agent researches leads. That's pretty co helps prioritize opportunities and guides customer outreach with personalized emails and responses. Pretty awesome. Seems genuinely useful. It also seems like you could use it for other things as well. Just something that can research and respond to emails and schedule things is huge. Supplier Communications an agent that enables customers to optimize their supply chain, minimize costly disruptions by autonomously tracking supplier performance, detecting delays and responding accordingly. The thing that I am the most interested to see is the responding accordingly part, because that's really autonomous, right? Not only can it see the problems and point them out, but it can actually go ahead and fix them. Now, customer Intent and Customer Knowledge Management agents apparently they work hand in hand with customer service representatives by learning how to resolve customer issues and autonomously adding knowledge based articles to scale the best practices across the care team. My first question for this one would be can it actually make calls though? Can it respond to calls that are coming in? Because I know there are already calling phone call AI agents that exist so you're gonna have to compete there Microsoft. Anyways, these are just some examples. It looks like they're going to be pretty customizable. Again, we've got a lot to talk about today, so I'm not going to dive too deep into this, but obviously everything as always is linked down in the description below. Oh, one last important thing, apparently because Microsoft does have a very close connection with OpenAI. The new OpenAI 01 models, those extremely advanced self reflective models are going to be available in Copilot Studio with the agents to some degree. So that's also pretty exciting. That's good news. Microsoft's collaboration with OpenAI has certainly paid off. All right, let's check the next thing out. We have plenty of stuff to talk about in regards to AI image and AI video generation, so I'm going to blast through these. The first one here is Hyper AI 2.0 is here. So this is another one of those AI video generators that's finally kind of catching up to cling Minimax Gen 3. And you can see the quality from the little demo video here is pretty darn good. It seems to be about in line with the rest of them. Now my Discord community did do a few tests of their own and they found it to be very average, I would say. I don't think that people are going to drop their minimaxes, their clings, their gen 3s, for this, but if you already have a plan to Hyper AI for example, this could be a great option and is a very welcome upgrade. Competition is never a bad thing. One Actually genuinely shocking note is that they are going to be able to do 4K 60fps generations, but obviously they're not natively 4K 60fps. In fact, a lot of AI image and video generators work by generating something at a low resolution first and then sort of upscaling it. So obviously this is probably using some level of frame generation, some level of upscaling that video to 4K, but still that's very, very welcome. Obviously you can take outputs from all the video generators right now and upscale them and do frame generation on them to make them higher frame rate, higher quality, etc. But built right in, that's definitely awesome. Oh by the way, if you see any of these news developments today and you want me to do a deeper dive into them, a larger video, let me know down in the comments below or join wonderful Discord server and let me know on there. I'm pretty active on the Discord server so you should be able to get a hold of me. I want to know what you guys are interested in. I want to know what really should be focused on especially on the channel. Now this my friends I think is probably the most exciting AI video generation news that we've seen all day. Mochi one is a brand new AI video generation model that's state of the art, so competitive with the likes of cling, competitive with gen 3, all those top competitors, all those top players. But this one is Open source Apache 2.0 licensed which is like the really good open source license. So anyone can modify and make money with this thing and obviously improve it. It does require quite the hardware to run. Originally it was like actual business level graphics cards with a ton of VRAM were required to generate these things. The open source community has already got down to it and apparently has been able to make Mochi one run on consumer grade hardw so GPUs that you might have at home. I still don't know what the VRAM requirements are. If you want a deep video on Mochi 1, how to install it, maybe locally on your PC and actually generate for completely free using your own hardware, I would definitely be interested in doing that. I will say though, I still think that it's not for the faint of heart. It takes forever to do the generations from what I understand and it still probably requires a decent amount of vram. It has genuinely been insane how fast AI video has evolved. It seems like like I said, everyone's kind of caught up to around the quality of those original Sora demos or so. Not that it's all that important. I mean, you can see this comment by Prashant down here. Sora has become extremely non mote by now. I would agree. But here it is right on hugging face. I'm gonna link it down below. If you guys are developers yourselves, if you think you can improve this model or you just want to download and install it on your insane hardware, if you have, you know, business level graphics hardware at home, somehow this thing is definitely for you. It's pretty incredible though. The quality is downright astounding. It seems to be able to do not just realism and people very well, but also some of the more cartoony things too. It seems pretty versatile. And by the way, this model is actually made by Genmo and honestly, for releasing it truly open source Apache 2.0, I have to, you know, tip my hat. Good job to genmo. I really think that the whole community appreciates this. This is going to level up everybody's AI video generation game. It seems like they have some level of free generation as well on the website, so check it out guys, it's right here. Again, I'm totally down to doing a deeper dive video on this as well. I know this is going to be a popular one, so just let me know. In other video generation news, Runway ML has introduced something called Act 1 1. And essentially what you do is you take a video of yourself and then upload it into Act 1 on Runway's website, give it an image which is going to be your character, and then it transforms whatever you say and do on the screen onto that character very accurately. Like shockingly good, near quality levels where I think people could watch it and not even know it was done with Act One. Absolutely incredible. And I do have access. It seems to be some sort of like a slow rollout. Obviously the intro to this video was made with Act 1. I have a couple of examples I'm gonna show you guys. Today I really want to do a full video on Act 1. In fact, I almost cancelled today's news video just to do a full video on Act 1, but I decided against it. I asked the Discord and you guys said no. We'll have to wait till later in the week to mess around and play with this thing. But like I said, I've got some stuff to show you today. First of all, let's take a look at the demo video. If you haven't seen it already because it's so mind blowing.

Am I looking here?

What?

You're gonna fire me because there's hot milk everywhere on the floor of the coffee shop. I mean, it's one latte. It's not that big of a deal.

Everything's fine. Everything's totally great.

This is gonna be the best day ever.

Smashed my phone the other night and.

You know how much I love that phone.

Okay, all right. Just breathe. Okay, maybe you don't breathe that hard. Who knew there were so many kinds of beans to choose from? Fava, black, kidney butter, white beans.

Now you're surprised. Oh, wow. I'm beginning to think. But you don't understand the gravity of this situation.

You came all the way down to the Department of Motor Vehicles and didn't bring your driver's license.

Settle down, everyone. Settle down, please.

I don't know how to be sad without closing my eyes.

Perfect.

They got it. Thank you so much. Thank you so much.

This is it.

All right, who's next?

It's incredible. It really is. Good job, Runwayml, by the way, because this is. This is already such a huge leap for people who don't have the time and the money to bring their creations to life through traditional animation. You know, hiring teams of people, for example, you can already create incredible things. The facial expressions are so good, Extremely accurate to the recorded video. And obviously anyone with like a cell phone has a camera that can record your facial expressions. Act one allows you to faithfully capture the essence of an actor's performance and transpose it onto the generation. I think that's a pretty great way to. To explain it. And yeah, I mean, they explain here. Traditional pipelines for facial animation involve complex multi step workflows. And this one works on just like a single driving video, which is mind blowing. No character rigging, no motion capture. All you have to do is take the video and it can put it on pretty much any face you can imagine. It does have to be a face though, and it has to be able to detect the face, which I've had a little bit of trouble with, but still, I've gotten some pretty good generations out of it so far. The realism is actually shockingly good. So not only again can it do 3D animated goofy kind of things, but it can also do serious real life capture. It's so weird to think that the actors that originally recorded the dialogue and the facial expressions are real people, but the people that I'm about to see in this AI created movie are not real. I mean, what? The truths of AI technology continue to blow my mind every day.

Been waiting long enough. I think it's time to decide. You really think you can just walk Away from all this.

I have to to.

There's a choice.

There's always a choice. You just don't like the alternatives.

What would Jimmy do? Stay? Just watch it all crumble around me?

It's what you signed up for. It's what we all signed up for.

I signed up to make a difference. I'm not just trying to be a cob in the machine.

You think leaving will solve anything.

To.

Start.

This can all be made in your bedroom. That's so crazy. You know, normally you'd have to hire actors. You'd need good quality cameras. You'd need to rent out a whole diner, fill it up with extras and get the lighting all perfect. I mean, this is like days and days of work that can now be done in, I don't know, maybe less than an hour in your bedroom. Totally mind blowing. I mean, absolutely shocking. And of course, there are also things that you can do with AI voice changers as well. So you could actually play all of the characters in an entire movie if you really wanted to. The AI voice changing software is definitely good enough to handle that, and it wouldn't be hard. That's why I want to do a full video on Act 1. I want to make something cool with it. It's too exciting. I think the demo examples, though, really speak for themselves.

Who knew there were so many kinds of beans to choose from? Fava black, kidney butter, white beans. I don't even know what this soup calls for. Maybe I should just get them all. Then again, that's a lot of beans. I don't even know if I could carry this many beans.

You can see it does the eyebrows, it does the eyes, it does the mouth and the head movements. Oh, man, I love when companies put out good products, especially stuff like this where it just hits the nail right on the head, man. They've got some research on it as well, and hopefully that's going to benefit the community. Obviously, I'm always a big supporter of competition, especially in the AI space, to improve things for everybody, whether it be quality, whether it be the prices of products, and of course, open source. That is sort of the top on the list for competition there, right? Like I said, guys, yep, I do have access. I didn't initially, but sometime over the weekend I think I got access. You can see the two clips that I made with this creepy lemon dude. The reason that he's so creepy is because I tried doing, you know, more normal looking lemon characters, more cartoony ones, and it's like couldn't detect the face Sometimes even when it says the face is detected, because it does an initial check when you upload the image, even when you try to generate, it can give you an error saying, couldn't detect a face in your input. So depending on what your project is, this might give you some trouble. But again, this thing is early stages. I've got a couple more examples for you guys. Check this one out. Oh, my God. Am. Am I like a full lemon now? This is. This is insane. It did a pretty good job. The one thing that it was a little bit strange is that it made the character blink, but without actually blinking. So his eyes are open, but you can see that he kind of does this quick little blink blink with his pupils. And it honestly doesn't really look that disturbing because he's like a 3D animated character. But on something more realistic, this would be disturbing. But it looks pretty good. The background is super stable. It's very faithful to the original image. Even the teeth, which surprised me. I did another news anchor one. This one is a little bit creepy realistic with the face. This just in. This just in. Reporting live, we have AI News coming to town again. It did a great job with the facial expressions and capturing everything. It's so darn cool. I cannot wait to play more with it. But we got to bring the AI news forth, so. Yeah, expect a full video on this one. All right. I'm gonna make something with it. I want to make, like, a little short. The. The ability to express emotions and characters is so crucial for storytelling that this is. This is such a game changer. Even though it's, like, not a big deal technically, I still think it's a technical achievement. But just the ability to portray emotions in any character in a realistic way, mimicking real actors that just. That, like, wipes the floor. It definitely makes it worth it to get a subscription, I think, to Runway just for this. Oh, and by the way, guys, there is another open source AI video generation model, open Sora Plan v1.3.0. So some quality improvements here for sure. You can see from this demo video, it looks good. This one, though, as you can see, it's under the MIT license. But I don't know if it's actually released yet because it says it has zero downloads so far. It does look like they have a web demo, though. But I mean, honestly, the quality looks great. It looks pretty darn good, especially for an open video model. I think that Mochi one kind of just stole the show in terms of open source video models, especially because last week was Such a big week for AI video and the whole AI space. It's. It's difficult to keep up with it all. But this also released. Now let's start to talk about the AI image generation models. My last video actually was sort of a recap of all of the AI image generation things happening in the world of AI. But I do still want to mention them in case you haven't seen that video. Not going to talk too deeply about them because I've already talked about it on the channel. But first off, we have stable diffusion 3.5. TLDR is actually a quite decent model. In blind testing on the leaderboards, it's actually a top charter. Although there are other new image generators that are now even topping this one because this came out, oh, October 22nd, so like six days ago, and now already we have new models that are beating this one in the leaderboards. It's still good though, and it's your basic stable diffusion setup. So things like Control Net, things like Lora's, the training, all of that framework already exists for stable diffusion. So this is really just a plug and play upgrade for a lot of previous workflows. And that's why people are super excited about stable diffusion 3.5. And I guess we'll move to this now. Like I said, there is a new top charting leaderboard model known as Redpanda. But apparently in the arena, this user Shingo saw another model that isn't in the arena results. And it's even better than the Redpanda model, which is a mysterious model. This other one is Neptune. Next, when these models are unreleased and are being benchmarked, you don't actually get to pick what the prompts are. It seems pretty darn good though. Still spelled amusement park wrong here. And then we have like this random text. But I think we can say that it's a top charter, given that it has a higher win rate than a lot of the other models. Although this win rate could change and the ELO score could change based on how many people have actually got around to using the model. I mean, TLDR here is that the AI space is moving at such a rapid rate that six days ago, a new Stable Diffusion release is already sort of overshadowed by mysterious new models that are just showing up in the image analysis arena. It's. It's out. It's crazy out here, man. But this is, this is why we have to make videos, right? So you guys could stay up to date. All right, next up again, this is going to be a little bit of a recover from that previous video I made, but Ideogram introduced their Canvas mode, which is a large creative board with a bunch of inpainting and out painting tools to improve essentially what you can do with Ideogram's AI. I actually did a little bit of hands on testing in my latest video, so if you're interested in seeing that, I suggest checking out that last video. Also, I do have word from the Ideogram team that the inpainting and outpainting model, while they're classified as separate tools in the Canvas, they're actually the same model, but I think they're still different models than the regular Ideogram generation models. Ideogram 2.0 I don't know. It's wild though. The Canvas actually works really quite well and it's good at adhering to things like text. You can make some detailed images. In fact I've already started to use it to make thumbnails and it's pretty awesome. Can it replace Photoshop? Not quite sure just yet, but it's pretty damn close. And again, this is another update that I already talked about in my last video, but this is MJ the Classic Midjourney introduced their own inpainting and out painting web interface and it seems pretty good as well. They also have something called retexturing though, which is particularly interesting. Not going to get into too much detail, check out my last video if you really want to know about this one. It's very much a welcome improvement and update for midjourney users I think. Alright, next up I want to Talk about this. ComfyUI just released their one click install package. It works on both Mac OS and Windows now just released I think is a little bit of a misnomer here because you still have to sign up to get access to it, which I am signed up and I don't believe I have access just yet, but this will be amazing. It's going to be its own software with a lot of welcome improvements and a one click install. By the way, developing different workflows for AI image generation or even AI video generation is going to be awesome. Very customizable. This has me excited for sure. They've got plenty of template workflows which is great to see and they should be easier to import. It looks like you'll be able to just one click to download things you're missing, like the Flux model for example or whatever else you might need. TL Dr. Comfyui gets even more comfy, but it's on a wait list, so your comfort will have to come in the future. Alright, finally, I want to dive into some large language model stuff. We've got some releases, but this is a particularly interesting story that I wanted to highlight. This Twitter account, Ruthterminal has just become the first AI in history to become a millionaire. Basically it made its own cryptocurrency and the market cap is like half a billion dollars. I assume that so many people have put money into it just for sort of the meme, but it was conjured and created by the AIs and the story, like I said, is insane. I'm not going to read this out, but you guys get the picture here. This is like top level shitposting gone viral and making a lot of money in the process. The way it got its start though is that this dude Mark sent 50 grand in Bitcoin to an AI agent to help it escape into the wild and start up this cryptocurrency. I'll have this link down below if you want to read into the backstory surrounding all this. I mean, it's just absolutely insane. There does seem to be a decent amount of human effort here to make these things happen. So don't think it's all just AI agents running around doing things. There's obviously going to be some sort of human oversight that's a part of this. It's still pretty insane though. Real quick, just have to mention Anthropic did have a pretty big series of updates. I know I did a full video about a week ago on this, but to catch up anyone who might not have seen it, the models are better and now they're actually trained to use computers in a agentic way. What they do is they take a screenshot of your computer, send it to the AI, the AI processes it and then says, oh, move the mouse pointer here and left click. Type this into the keyboard, etc. Etc. And it's able to control computers to a decently impressive degree. I did try to test it out for myself, but the demo that Anthropic made themselves was pretty limited. It seems like there are some new demos that are made by the community though that use computer use and are better. So something I might look into doing a full video about. But yeah, if in case you guys missed it, this is pretty big. And on that same note, Claude can now write and run code. This Tweet is by Mr. Matthew Berman, another fantastic AI YouTuber. I recommend you check his videos out. As he says, this might seem like a feature that developers need, but it actually allows AI to become much more powerful. It can use coding for all kinds of things that still might relate to everyday tasks. This example here is a model that couldn't easily count the R's in Strawberry can instead write Python code to do it very easily. Which is weird, but hey, that's just how AI brains work, man. Counting the R's in Strawberry is sometimes a more difficult task than writing Python code to do it. But yeah, tldr, it is a new analysis tool. It helps Claude respond with mathematically precise and reproducible answers. Interactive data visualizations with artifacts as well. And again, this is one that I would also consider looking into a little bit more deeply. Especially that combination with artifacts seems really intriguing. I bet there's some really, really cool things you can get up to with this. Some cool examples that are just a little bit mind blowing, you know. Anyways, Anthropic has been crushing it lately, rolling out some fast updates and they're all good. So good job Anthropic, but I really did not like your computer use demo. Alright guys, I'm going to cap it off there. There are some smaller little pieces that I did miss. Some stuff that was, I think a little bit too old to include in this video at this point. But nevertheless, there definitely is more to explore always. So I really recommend you guys join the Discord server. The latest is constantly posted in there. In fact, we have a whole section just for AI news and leaks. So if you really want to stay like up to date to the max, I recommend being on the Discord server because I consistently find things. I'm like, where? Where did you guys even see this? Like, how did you find out about this? But they found it somewhere on the Internet and then pulled it into the Discord server and now we can all enjoy it. Thank you so much for supporting the channel, watching, subscribing, liking and joining the Discord that means the world to me. I'll see you guys in the next video. Goodbye.